This document contains results for comparing row and column sampling for consensus partitioning on the five datasets ( Golub leukemia dataset, HSMM single cell RNASeq dataset, MCF10CA single cell RNASeq dataset, Ritz ALL dataset and TCGA GBM microarray dataset). For each dataset, four consensus partitioning methods (SD:hclust, SD:skmeans, ATC:hclust and ATC:skmeans) were applied, and each method ran for 100 times so that the variability of 1-PAC can be captured. The random sampling was done by rows and by columns. Each individual cola run was done with default parameters. The scripts for the analysis can be found here.
For each dataset, there are four plots:
- boxplots that show the distributions of 1-PAC scores at each k (number of subgroups) for each method.
- mean difference of the 1-PAC score between row-sampling and column-sampling.
- heatmaps that directly show the partitions from 100 runs. Each row corresponds to one cola run and the color in the heatmap only corresponds to the subgroup labels, while not the stability of the partitioning in that run.
- barplots that show the concordance of the partitions in 100 runs for the row-sampling or for the column-sampling separately, as well as the concordance between row-sampling and column-sampling. Note the scale on y-axes is transformed as \(1 - \sqrt{1-y}\).
Golub leukemia dataset
Figure 1. Distribution of 1-PAC scores.
Figure 2. Mean difference of 1-PAC between row-sampling and column-sampling.
Figure 3. Individual partitions from row-sampling or column-sampling.
Figure 4. Concordance of the partitioning by row-sampling or/and column-sampling.
HSMM single cell RNASeq dataset
Figure 5. Distribution of 1-PAC scores.
Figure 6. Mean difference of 1-PAC between row-sampling and column-sampling.
Figure 7. Individual partitions from row-sampling or column-sampling.
Figure 8. Concordance of the partitioning by row-sampling or/and column-sampling.
MCF10CA single cell RNASeq dataset
Figure 9. Distribution of 1-PAC scores.
Figure 10. Mean difference of 1-PAC between row-sampling and column-sampling.
Figure 11. Individual partitions from row-sampling or column-sampling.
Figure 12. Concordance of the partitioning by row-sampling or/and column-sampling.
Ritz ALL dataset
Figure 13. Distribution of 1-PAC scores.
Figure 14. Mean difference of 1-PAC between row-sampling and column-sampling.
Figure 15. Individual partitions from row-sampling or column-sampling.
Figure 16. Concordance of the partitioning by row-sampling or/and column-sampling.
TCGA GBM microarray dataset
Figure 17. Distribution of 1-PAC scores.
Figure 18. Mean difference of 1-PAC between row-sampling and column-sampling.
Figure 19. Individual partitions from row-sampling or column-sampling.
Figure 20. Concordance of the partitioning by row-sampling or/and column-sampling.